[PATCH] tree-optimization/98855 - fix some vectorizer cost issues

[Richard Biener]

This fixes us not costing vectorized bswap for SLP as well as
avoiding biasing to the vectorized side when costing single-argument
PHIs.  Instead we assume coalescing here and cost them with zero cost
for both the scalar and vectorized code.

This doesn't fix the PR on its own.

Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.

2021-02-04  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/98855
	* tree-vect-loop.c (vectorizable_phi): Do not cost
	single-argument PHIs.
	* tree-vect-slp.c (vect_bb_slp_scalar_cost): Likewise.
	* tree-vect-stmts.c (vectorizable_bswap): Also perform
	costing for SLP operation.
---
 gcc/tree-vect-loop.c  |  8 ++++++--
 gcc/tree-vect-slp.c   |  7 +++++++
 gcc/tree-vect-stmts.c | 13 ++++++-------
 3 files changed, 19 insertions(+), 9 deletions(-)

diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
index 200ed27b32e..f1f16e3b1b9 100644
--- a/gcc/tree-vect-loop.c
+++ b/gcc/tree-vect-loop.c
@@ -7777,8 +7777,12 @@ vectorizable_phi (vec_info *,
 			       "incompatible vector types for invariants\n");
 	    return false;
 	  }
-      record_stmt_cost (cost_vec, SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node),
-			vector_stmt, stmt_info, vectype, 0, vect_body);
+      /* For single-argument PHIs assume coalescing which means zero cost
+	 for the scalar and the vector PHIs.  This avoids artificially
+	 favoring the vector path (but may pessimize it in some cases).  */
+      if (gimple_phi_num_args (as_a <gphi *> (stmt_info->stmt)) > 1)
+	record_stmt_cost (cost_vec, SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node),
+			  vector_stmt, stmt_info, vectype, 0, vect_body);
       STMT_VINFO_TYPE (stmt_info) = phi_info_type;
       return true;
     }
diff --git a/gcc/tree-vect-slp.c b/gcc/tree-vect-slp.c
index 10b876ff5ed..2305bbdec3a 100644
--- a/gcc/tree-vect-slp.c
+++ b/gcc/tree-vect-slp.c
@@ -4297,6 +4297,13 @@ vect_bb_slp_scalar_cost (vec_info *vinfo,
 	}
       else if (vect_nop_conversion_p (orig_stmt_info))
 	continue;
+      /* For single-argument PHIs assume coalescing which means zero cost
+	 for the scalar and the vector PHIs.  This avoids artificially
+	 favoring the vector path (but may pessimize it in some cases).  */
+      else if (is_a <gphi *> (orig_stmt_info->stmt)
+	       && gimple_phi_num_args
+		    (as_a <gphi *> (orig_stmt_info->stmt)) == 1)
+	continue;
       else
 	kind = scalar_stmt;
       record_stmt_cost (cost_vec, 1, kind, orig_stmt_info,
diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index 5eb7b2d1d6e..0bc1cb1c5b4 100644
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -3002,13 +3002,12 @@ vectorizable_bswap (vec_info *vinfo,
 
       STMT_VINFO_TYPE (stmt_info) = call_vec_info_type;
       DUMP_VECT_SCOPE ("vectorizable_bswap");
-      if (! slp_node)
-	{
-	  record_stmt_cost (cost_vec,
-			    1, vector_stmt, stmt_info, 0, vect_prologue);
-	  record_stmt_cost (cost_vec,
-			    ncopies, vec_perm, stmt_info, 0, vect_body);
-	}
+      record_stmt_cost (cost_vec,
+			1, vector_stmt, stmt_info, 0, vect_prologue);
+      record_stmt_cost (cost_vec,
+			slp_node
+			? SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node) : ncopies,
+			vec_perm, stmt_info, 0, vect_body);
       return true;
     }
 
-- 
2.26.2